remote sensing reflectance and derived products: modis …...modisa instrument calibration update...
TRANSCRIPT
Remote Sensing Reflectance and Derived Products: MODIS and VIIRS
MODIS/VIIRS Science Team, 15-19 October 2018, Silver Spring, MD
Bryan FranzOcean Ecology Laboratory
NASA Goddard Space Flight Center
Project Team
MODIS Algorithm MaintenanceRemote Sensing Reflectance, Chlorophyll, Diffuse Attenuation
Science of Terra/Aqua/SNPPAdvancingtheQualityandContinuityofMarineRemoteSensingReflectanceandDerivedOceanColorProductsfromMODIStoVIIRS
TeamMember RoleBryanFranz(PI) ProjectleadAmirIbrahim(Co-I) AtmosphericcorrectionSeanBailey(Co-I) VicariouscalibrationGerhardMeister(Co-I) InstrumentcalibrationJeremyWerdell (Co-I) Bio-opticalalgorithmsZiaAhmad(Supp) AC&radiativetransfercodeGeneEplee (Supp) Instrumentcalibration(VIIRS)Shihyan Lee(Supp) Instrumentcalibration(MODIS)ChrisProctor(Supp) InsituvalidationErdem Karakoylu (Supp) Uncertainties
andtheOceanBiologyProcessingGroup
Contents
1. StatusofMODIS&VIIRSOceanColorProducts
2. StatusandPlansforAlgorithmAdvancement
1. Multi-bandAtmosphericCorrection
2. Rrs UncertaintyProduct
Ocean Color Reprocessing R2018.0
Completed• Dec2017(VIIRS/SNPP),January2018(MODIS/Aqua),April2018(MODIS/Terra)
Purpose• incorporateupdatestovicariouscalibrationduetorevisedMOBYtime-series• incorporateupdatestoinstrumentcalibration• noalgorithmchanges
Impact of MOBY Revisions on MODISA Ocean ColorGlobal Deep Water Time-Series Test (MOBY change only)
MOBYinsituradiometryisusedforvicariouscal ofvisiblebands,allmissions
MOBYreprocessedbyMOTin2016and2017• reviseddepthofsensorarms• newstraylightcorrection
Impactwasto• increaseRrs by2-10%• decreaseglobalChl by8%
RrsR
atio
Rrs(sr
-1)
ChlRatio
Chl(mgm
-3)
MOBYbiasshift+2-10%inRrs
MOBYbiasshift-8%inChl
MODISA Instrument Calibration UpdateGoalwastomitigatesometemporalartifactsobservedinpreviousoceancolortrends,believedtobeassociatedwithcal datasmoothing/fittingdecisions.
Completeend-to-endre-analysisofon-boardcalibrationwasperformed,largelyfollowingMCSTapproach,butwithsomedifferences.• SD/SDSMscreens,SDBRF:derivedfromsimplefitofyawmaneuverdata,notgeometricmodeling
• SDSMCal:SDdegradationsinred/NIRweredeterminedbywavelengthmodeling(Leeetal.,2018)
• RVS:simpleatmosphericcorrectionaddedwhencomputingdeserttrendstotrackRVS in412,443nm
• ModulatedRSR:impactonoceandataestimatedandusedtoadjust412nmtemporalgains(Leeet.al.2017)
• RSBLUTs:useofsmoothingratherthanfittingtotheinstrumenttemporalcalibrationtrendingand
characterization(i.e.,noassumptiononfunctionalform)
Monthlyupdatetomaintaincalibrationtrendconsistencyinforwardprocessing.
MODISA Global Deep-Water (R2018.0 vs R2014.0)
RrsR
atio
Rrs(sr
-1)
ChlRatio
Chl(mgm
-3)
instrumentcalibrationupdate
MOBYbiasshift+2-10%inRrs
MOBYbiasshift-8%inChl
Late-missiondriftinblueRrs resolved.
Largeseasonalspikesinchlorophyllremoved.
MODISA Rrs Anomaly Trend in Global Oligotrophic Water
Inveryclear-water,openoceanregions(Chl <0.1),weexpectminimalvariabilityingreenspectralbands.
R2018showsimprovedstabilityintemporaltrendsofwater-leavingreflectanceinthegreenbandsofMODIS/Aqua.
R2014.0
R2014.0
R2018.0
R2018.0R
rs(5
47)
Rrs
(555
)
instrumentcalibrationupdate
biasshift+3-5%inRrs
biasshift-9%inChl
RrsR
atio
Rrs(sr
-1)
ChlRatio
Chl(mgm
-3)
VIIRS/SNPP Global Deep-Water (R2018.0 vs R2014.0)
InstrumentCalibrationupdatesinclude:
• revisedtemporalcalwithadditionalsolar&lunarmeasurementssinceR2014
• lunarmeasurementscorrectedfordetectorgainvariability
• absolutecalibrationofallspectralbandsbasedonobservationsofthesunthroughthesolardiffuser
VIIRS/SNPP R2018.0 Reprocessingimproved agreement with in situ
Rrs(412) Rrs(443) Chlorophyll
R2014.0
R2018.0
Bias
Bias=5%Bias=-0.00032 Bias=+0.00001
VIIRS/SNPP R2018.0 Reprocessingimproved agreement with in situ
Rrs(412) Rrs(443) Chlorophyll
R201
4.0
R201
8.0
Bias
Bias = 5%Bias = -0.00032 Bias = +0.00001
Bias=-0.001
MODISA R2018.0 Reprocessingimproved agreement with in situ
R2014.0
R2018.0
Rrs(412) Rrs(443)
Bias
Chlorophyll
Bias=32%
Bias=18%
Bias=-0.00077
Bias=+0.00005 Bias=+0.00008
Bias=-0.00027
In situ Rrs Match-up Statistics for MODISA
Product NumberMatchups
MeanBias(sr-1) MeanAbsoluteError(sr-1)
R2014.0 R2018.0 R2014.0 R2018.0Rrs(412) 3934 -0.00077 0.00005 0.00125 0.00099
Rrs(443) 4134 -0.00027 0.00008 0.00082 0.00075
Rrs(488) 3772 -0.00067 -0.00050 0.00088 0.00077
Rrs(531) 2076 -0.00063 -0.00057 0.00083 0.00079
Rrs(547) 3619 -0.00052 -0.00048 0.00078 0.00076
Rrs(555) 3534 -0.00081 -0.00075 0.00096 0.00090
Rrs(667) 3573 -0.00017 -0.00016 0.00030 0.00029
Rrs(678) 472 -0.00016 -0.00014 0.00034 0.00033
• equalorreducedmeanbiasandMAEinallbands• largenegativebiasinbluereducedtosmallpositivebias
identicalfielddatasourcesfromAERONET-OCandSeaBASS
0.10
0.12
0.14
0.16
0.18
0.20
Chla
(mg
m−3)
0.10
0.12
0.14
0.16
0.18
0.20
Chla
(mg
m−3)
Jan
1998
Jan
2002
Jan
2006
Jan
2010
Jan
2014
−20
0
20
40
Chla
(%)
−0.04
−0.02
0.00
0.02
0.04Ch
la A
nom
aly
(mg
m−3)
−0.04
−0.02
0.00
0.02
0.04Ch
la A
nom
aly
(mg
m−3)
Jan
1998
Jan
2002
Jan
2006
Jan
2010
Jan
2014
−20
−10
0
10
20
Chla
Ano
mal
y (%
)
a.
b.
SeaWiFS MODISA VIIRS
Multivariate Enso Index (MEI)
SeaWiFS MODISA VIIRS
21-year Multi-mission Chlorophyll TimeseriesDeep-Water, +/- 40 deg latitude
• MODISAingoodagreementwithSeaWiFS andVIIRS(through~2015)
• anomalytrendsgenerallyconsistentwithexpectationsbasedonMEI
• suspectlatemissiontrendinVIIRS(didnotrecoverfrom2015/16ENSOevent)
• likelyissueisunresolvedimpactsinbluebandstochangingspectralresponseofVIIRS(TBD)
Chloroph
yll(mgm
-3)
Chloroph
yllA
nomaly(m
gm
-3)
from Ibrahim et al., Ocean Optics XXIV, October 2018, Dubrovnik, Croatia.
Gordon and Wang, 1994
MSE MBAC
SSàMS
AC Algorithm Refinement
TOAradiance
Lt
Rayleigh,gas,andglintcorrection
Estimate the optical depth at the longest high performance band
Estimate the aerosol spectral reflectance for a set of models
Lrgc
𝛁𝝌𝟐minimization
Averagebestmodels,calculatetransmittances
Rrs ,𝝉a,Angstrom
Multi-band Atmospheric Correction (MBAC)
MBAC Cost Function
observed/predictedaerosolreflectance adaptivespectralweight
radiometricuncertainty
SensorWavelength(l,nm)
MODIS 748 859 869 1240 1640 2130
VIIRS 746 868 1238 1604 2258
PACE-OCI 750 860 870 1038 1250 1615 2130 2260
𝑆𝑊(𝜆)
0
1
NIRiterations(Baileyetal.,2010)increasingwatercomplexity
Operational MBACChesapeake Bay
Pamlico Sound
Operational MBAC
• MBACchlorophyllretrievalssimilartostandardalgorithm,exceptinturbidcoastalregions.
• Differenceisduetoretrievedaerosolmodels(aerosoltype).
Application to MODISA (MBAC vs Std AC)
MODISA In Situ Match-up Analysis (MBAC vs Std AC)
SeaBASS in-situ matchups to
water-leaving reflectance shows
reduction in error using MBAC
relative to standard algorithm, at
all spectral bands.
Artifacts ?
Highly turbid example for upper
Chesapeake Bay, MBAC shows
more meaningful (positive) Rrs
retrievals in blue channels.
• Algorithm
• Sensor Random Noise
• Sensor Systematic Noiseuncertainty
𝐿+
RetrievalAlgorithm(InverseProblem)
uncertainty
bias
𝑅-.
observations retrievals
Development of a Per-Pixel Rrs(l) Uncertainty Product
from Ibrahim et al., Ocean Optics XXIV, October 2018, Dubrovnik, Croatia.
Uncertainty per-pixel
TOA measurements noise
Step-by-steperrorpropagation
Error Propagation
Sensor Random Noise
• SensornoisecanbepropagatedthroughtheMSE/MBACalgorithmsusingstandarderrorpropagation.
• ResultsagreeverywellwithMCsimulations(presentedpreviously).
MC Simulation
𝜃0123
𝜃.45 = 10° 𝜃.45 = 20°
𝜃.45 = 30° 𝜃.45 = 60°
𝜃0123
𝜃0123
𝜃0123
Simulationparameters:• 𝜃0123 = 0°: 80°,• ∆𝜑 = 0°: 180°,• 𝜃.45 = 10°, 20°, 30°, 60°• 𝜏D = 0.05: 0.35• RH = 77.5% à Not in the LUT• 𝑓 = 0: 0.95• chlor-a=0.03mg/m3
• total#cases:1037400
• average∆𝜌3 443 =0.0015
• uncertaintyincreaseswithsolar
zenithangle
• uncertaintyincreasessignificantly
forlargerfine-modefractioncases
duetothelinear-mixinglimitations
𝝀 = 𝟒𝟒𝟑𝒏𝒎
Algorithm Uncertainty
𝜃0123
𝑅𝑒𝑑𝑢𝑐𝑡𝑖𝑜𝑛𝑖𝑛∆𝜌3𝑤𝑖𝑡ℎ𝑀𝐵𝐴𝐶
𝜃0123 𝜃0123
Simulationparameters:• classicalcovarianceanalysis• assumecorrelatedbiasof2%onbands>750nmand0.5%onVNIRbands
• assume10%correlatedcovariancebetweenbands
average uncertainty:
• Operational 2-band AC à 0.0036
• MODIS-MBAC 6-band à 0.0024
• Uncertainty is reduced by à 34%
𝝀 = 𝟒𝟒𝟑𝒏𝒎, 𝜽𝒔𝒖𝒏 = 𝟑𝟎°
𝑀𝐵𝐴𝐶 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑𝐴𝑙𝑔𝑜𝑟𝑖𝑡ℎ𝑚
Sensor Systematic Error
Future DevelopmentMulti-BandAtmosphericCorrection• assess/improvespectral-relativecalibrationforNIR/SWIRregion(MODIS&VIIRS)• assess/improveperformanceofadaptiveweightingscheme• initiateglobal-scaletestingforMODIS&VIIRS
Rrs UncertaintyProduct• completethealgorithmframeworkforRrs uncertaintyproduct(combined
random/systematic/algorithmerror)• refinesystematicerrortermsforeachinstrument• assessuncertaintyestimatesrelativetomatch-upresults(closureexperiment)
InstrumentCalibration• reassessVIIRSspectralresponsechanges,impactontrendsinblue• evaluateMODIS-TerraR2018.0results